Identifying Decision Points for Safe and Interpretable Reinforcement Learning in Hypotension Treatment
This work provides a method for identifying key decision points in continuous medical trajectories, which could benefit clinicians by making RL recommendations more interpretable and efficient for patient care.
This paper addresses the problem of fixed-interval time discretization in batch reinforcement learning for healthcare by developing a framework to identify critical decision points in continuous trajectories. This approach leads to a reduced state space, enabling faster planning and easier inspection by clinical experts in the context of hypotension treatment.
Many batch RL health applications first discretize time into fixed intervals. However, this discretization both loses resolution and forces a policy computation at each (potentially fine) interval. In this work, we develop a novel framework to compress continuous trajectories into a few, interpretable decision points --places where the batch data support multiple alternatives. We apply our approach to create recommendations from a cohort of hypotensive patients dataset. Our reduced state space results in faster planning and allows easy inspection by a clinical expert.